Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 379 - Bias and Interpretability in Biometrics for Forensic Science
Type: Topic-Contributed
Date/Time: Thursday, August 12, 2021 : 12:00 PM to 1:50 PM
Sponsor: Section on Statistics in Defense and National Security
Abstract #317637
Title: Machine Testimony
Author(s): Andrea Roth*
Companies: UC Berkeley School of Law
Keywords: law; jury; expert witness; DNA; forensic evidence
Abstract:

This presentation will offer a legal expert’s overview, for statisticians, of the use of “black box” proprietary algorithms at all stages of criminal justice, from investigation (predictive policing, use of AI to determine falsity of confessions, etc.) to pretrial detention (risk assessment) to proof of guilt/innocence at trial (from probabilistic genotyping software to text-based authorship attribution to facial recognition software), to legal informatics (e.g. “moneyball prosecution”). The presentation will discuss legal issues arising from such proof, including inaccurate verdicts based on unreliable results (discussing how concepts of “hearsay” and “confrontation” might apply to machines, and discussing “Daubert”/”Frye” expert requirements and authentication rules), and pros/cons of reforms such as source code disclosure. The author will also discuss intellectual property and privacy issues that arise with respect to proprietary algorithms, and, time permitting, several other recurring legal issues of likely general interest to statisticians, including quantification of standards of proof, reluctance to allow Bayesian reasoning, and juror understanding of LRs.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2021 program